Personal project · Applied AI

Answers with evidence, not just text.

Semantic search over Mexican standards, with citations, evaluation and per-query traces.

Backend, retrieval and evaluation · Web demo

Python
FastAPI
ChromaDB
Pytest

Open app View code

NOM RAG interface with a consultation assistant, suggested questions and a sources panel on a dark background.

The problem

Retrieve passages from standards and generate answers that can be traced back to the source.

My contribution

I developed the RAG pipeline, API and a retrieval-and-answer evaluation suite.

Technical decisions

Clause-based parsing, multilingual E5 embeddings and persistent vector search in ChromaDB.
Follow-up query reformulation, citations and controls for out-of-corpus questions.
Pytest evaluation and validated JSON verdicts; retrieval, token and latency traces.
Docker Compose and GitHub Actions for execution and evaluations.

Scope and limitations

A personal project with a limited corpus. Answers should be checked against their sources and do not replace legal advice; evaluations cover only the included test cases.

Next projectPouV2